Melpomene-70B-0307-Uncensored: A Hybrid Thinking Model

Melpomene-70B-0307-Uncensored is a 70 billion parameter language model developed by Nabbers1999, built upon a base of unsloth/Llama-3.3-70B-Instruct. This model is a unique merge, created using the DARE TIES method, combining the strengths of several prominent models to achieve a distinct profile.

Key Capabilities & Characteristics

• Hybrid Intelligence: Merges the creative writing prowess of Lumimaid-v0.2-70B and Strawberrylemonade-L3-70B-v1.1 with the deep reasoning abilities of DeepSeek-R1-Distill-Llama-70B.
• Uncensored Output: Explicitly designed as an "abliterated via orthogonalization and reinforced via direction-only DoRA training" version of its predecessor, Thalia, indicating a focus on unfiltered responses.
• Advanced Thinking: Optimized for complex thought processes, though it requires the prefilling of an <think> tag for proper function, as its ancestry can lead to un-tagged internal monologues.
• Merge Composition: Constructed from NeverSleep/Lumimaid-v0.2-70B, deepseek-ai/DeepSeek-R1-Distill-Llama-70B, and sophosympatheia/Strawberrylemonade-L3-70B-v1.1, with specific density and weight parameters for each component.

Ideal Use Cases

This model is particularly suited for applications requiring:

• Creative Content Generation: Leveraging the writing capabilities inherited from its chat model parents.
• Complex Reasoning Tasks: Benefiting from the deep reasoning foundation provided by Deepseek.
• Unfiltered Dialogue: For scenarios where uncensored or less constrained responses are desired.
• Experimental AI Development: Given its unique merging and training methodology, it's suitable for developers exploring hybrid model behaviors and advanced prompting techniques.